Deep Learning with PyTorch

5-D tensors

One common example where you may have to use a five-dimensional tensor is video data. Videos can be split into frames, for example, a 30-second video containing a panda playing with a ball may contain 30 frames, which could be represented as a tensor of shape (1 x 30 x 224 x 224 x 3). A batch of such videos can be represented as tensors of shape (32 x 30 x 224 x 224 x 3)—30 in the example represents, number of frames in that single video clip, where 32 represents the number of such video clips.